Carnegie Mellon University
18-849b Dependable Embedded
Authors: Chris Inacio
Embedded systems are complex designs often involving many disciplines in
order to create achieve the market demands of price, performance, reliability,
and functionality. As part of this multi-disciplinary design, the
reliability of the mechanical components must be taken into account for
most of the systems market objectives. Mechanical reliability has
its foundations based in material science, tribology, and deformation mechanics.
Further, understanding of statistics and probability is paramount to understanding
and creating a reliable mechanical system. While the statistics and
probability are discussed elsewhere, introductory materials on tribology
and some material science are introduced. Also covered here are precompiled
data sources and data modeling techniques relevant to mechanical reliability.
Mechanical reliability is a very old subject, for as long as man has built
things, he has wanted to make them as reliable as possible. It has
been possible in the past to make reliable mechanical systems by simply
over engineering them by large factors in order to avoid knowledge of the
materials, modes of failure, and other factors which cause mechanical systems
Today, mechanical reliability is a vastly different topic then when
man first started to build structures like bridges and houses. The
study and practice of mechanical reliability is a diversified field.
Research today is often based in material science or tribology (the study
of lubricants and wear or fatigue.) The material science knowledge
has greatly advanced the state of the art in mechanical reliability, especially
in the past fifty years. The combination of advanced materials and
the statistical modeling of components has lead to a stratified approach
to predicting the reliability of mechanical components.
Many standard mechanical components, ball bearings, roller bearings,
guide pins, control valves, etc. can be well predicted using historical
data. Many producers of standard components along with the military
have generated large databases of stresses, parts, and the the corresponding
reliability data for standard parts. The availability of this data,
while generally conservative since the science of building of the components
is advancing faster than the historical data gets revised, is a valuable
tool in estimating the reliability of standard components. While
standard components have large amounts of historical data which aides in
predicting their reliability, predicting the reliability of custom components
is less precise.
Mechanical components which are design and fabricated for a specific
system cannot easily have their reliability predicted. The variables
effecting the reliability include manufacturing variance, material variance,
and applied stress, to name a few. There are some models developed
in order to predict the reliability of these components, but nothing is
as accurate as knowing the history of a specific item being used in the
Part of generating the valuable historical data to predict future reliability
of mechanical components is classifying their failure. The Reliability
Analysis Center (RAC) developed a classification for the failure of mechanical
components which includes the cause, mode, and mechanism. This data,
while useful for predicting future reliability is also useful for developing
regular maintenance schedules and improving the design of similar systems.
Mechanical failure is classified by three different properties, which together
serve to describe a failure in complete detail. The three key classifications
of mechanical failure are the mechanisms, cause, and mode.
These three items combine to give the engineer a key view in understanding
how and why a part failed and what can be done to prevent a failure in
the future. It is imperative to understand that mechanical parts,
like most other items, do not survive indefinitely without maintenance.
A large portion of mechanical reliability is determining when maintenance
should be done in order to prevent a failure.
Available tools, techniques, and metrics
Mechanical engineering is a very old practice. Man has desired to
make mechanical devices for ages. In man's pursuit of building tools,
he has refined the art and craft of designing mechanical tools. Today,
there are various tools available for mechanical engineers. Systems,
such as finite element analysis are very powerful. In addition, the
technology to test mechanical components is impressive. In man's
long pursuit of building mechanical tools, he has developed an advanced
and complex set of mature tools, including simulation and testing
to aide in the design process.
The best metrics, for standard or custom components, is historical data
of that component in the system in which it will be used. For example,
the best data on when the maintenance for a roller bearing in a conveyer
line should be replaced, is knowing when the roller bearings in the same
conveyer line needed replacing previously. Obviously, for new systems,
this is not possible. If the new system is similar enough to a system
that the designer has the reliability data, that data should be used.
For completely new systems, however, alternative means of estimating reliability
must be employed.
Many standard components have a long history in mechanical engineering.
These standard components are used in many machines in many places without
any customization. Such components, like ball bearings, are available
from a multitude of vendors. Since the use of these components is
so wide spread, it makes sense to maintain reliability information on them.
The following table contains a list of references with information that
may be applicable to various standard mechanical components.
Comparing the first three in the table leads to some interesting results.
GIDEP contains data that is submitted by the manufacturers and it is not
uncommon for components, specified almost exactly the same to vary in failures
rates by many orders of magnitude. Unfortunately, the US government
does not have the resources to regulate the data entered into the database.
GIDEP is, however, very useful for getting a rough estimate of reliability
early in the design process. NPRD-3 reliability data is taken from
actual usage of those components in military equipment. For this
reason, the data in the NPRD-3 is very good, however, all of the data is
listed as failures per million hours. Unfortunately, for cyclical
equipment, this is a very poor measure. The appropriate term for
cyclical equipment would be in failures per million cycles. Lastly,
the AVCO data contains failure data information for usage in various environments.
Failure for various components can vary greatly depending on the environment
in which they are used. The problem with the AVCO data is that it
is from 1962. Technology, especially materials technology in this
case, have advanced tremendously since 1962.
||Publisher and Date
||Government Industry Data Exchange Program
||United States Department of Commerce
||Nonelectronic Parts Reliability Data
||Reliability Analysis Center, RAC, New York, 1985
||D. R. Earles, AVCO Corp., 1962
||Guidelines for Process Equipment Reliability
||American Institute of Chemical Engineers, 1990
|Davenport and Warwick
||A further Study of Pressure Vessels in the UK 1983-1988
||AEA Technology - Safety and Reliability Directorate, 1991
|DEFSTAN 0041, Part 3
||MOD Practices and Procedures for Reliability and Maintainability,
Part 3, Reliability Prediction
||Ministry of Defense, 1983
|R. F. de la Mare
||Pipeline Reliability; report 80-0572
||Det Norske Veritas/Bradford University, 1980
|Dexter and Perkins
||Component Failure Rate Data with Potential Applicability to a Nuclear
Fuel Reprocessing Plant, report DP-1633
||E. I. Du Pont de Nemours and Company, USA, 1982
||European Industry Reliability Data Handbook, Vol. 1 Electrical Power
||EUORSTAT, Paris, 1991
|ENI Data Book
||ENI Reliability Data Bank - Component Reliability Handbook
||Ente Nazionale Indocarburi (ENI), Milan, 1982
|Green and Bourne
||Wiley Interscience, London, 1972
|IAEA TECDOC 478
||Component Reliability Data for Use in Probabilistic Safety Assessment
||International Atomic Energy Agency, Vienna, 1998
|IEEE Std 500-1984
||IEEE Guide to the Collection and Presentation of electrical, Electronic
Sensing Component and Mechanical Equipment Reliability Data for Nuclear
Power Generating Stations
||Institution of Electrical and Electronic Engineers, New York, 1983
|F. P. Lees
||Loss Prevention in the Process Indestries
||Butterworth, London 1980
|MIL - HDBK 217E
||Military Handbook - Reliability Prediction of Electronic Equipment,
||US Department of Defense, 1986
||Offshore Reliability Data (OERDA) Handbook
||OERDA, Hovik, Norway, 1984
||Offshore Reliability Data, 2nd Edition
||DnV Technica, Norway, 1992
||Reliability Data Book for Components in Swedish Nuclear Power Plants
||RKS - Nuclear Safety Board of the Swedish Utilities and SKI - Swedish
Nuclear Power Inspectorate, 1987
|H. A. Rothbart
||Mechanical Design and Systems Handbook
|D. J. Smith
||Reliability and Maintainability in Perspective
||Macmillan, London, 1985
|Smith and Warwick
||A Survey of Defects in Pressure Vessels in the UK (1962 - 1978) and
its Relevance to Primary Circuits, report SRD R203
||AEA Technology - Safety and Reliability Directorate, 1981
Data Sources for Part Reliability
||Reactor Safety Study. An Assessment of Accident Risks an US Commercial
Nuclear Power Plants, Appendix III, Failure Data
||US Atomic Energy Commission, 1974
While none of the data sources provides the perfect data for use.
The data contained in the various generic sources may not be perfect, however,
it is still very useful for creating new designs. In order to estimate
the reliability of a new mechanical design, it is necessary to use some
estimates. Generally, the estimates given for the generic data is
conservative enough that it can be used safely without concern that the
device will fail significantly more than estimated. The downside
to the conservative data is that it may cause the designer to increase
cost in order to increase reliability.
One could capture just failure rate data for custom mechanical components.
There, is, of course, much more information that can be captured and used
to improve both the maintenance and future designs of the system.
It is common in engineering, especially in reliability engineering to quantify
and classify failures of systems; unfortunately, the terminology is not
the same. For classifying mechanical failures we will be using the
RAC terminology. The failures are categorized into three fields,
the mechanism of failure, the cause of failure, and the mode of failure.
Mechanisms of Failure
The cause of failure can be something as simple as a loss of lubrication.
The mode of failure is the result of the failure mechanism. An example
classification of a mechanical failure is a roller bearing that experiences
distortion, (the mechanism of failure,) due to a loss of lubrication, (the
cause of the failure,) and caused excessive vibration, (the mode of failure.)
The mode of failure can be used as an aide in diagnosis of system failures.
Further, there are two insights which can can be gained by classifying
failures for the design engineer: the practicing engineer can gain a deeper
understanding of the stresses present in a mechanical system; secondly,
by classifying mechanical failures, an engineer can understand better which
analyses need to be done in order to better predict the reliability of
the component in its environment.
A simple example of using generic data to estimate the reliability and
determine the probability for a system to complete its mission follows.
This example is completely fictional and from [Ireson].
If the mission time for this assembly is 500 hours, the probability of
success would be:
||General failure rate per million hours
||Total failure rate per million hours
|Heavy-duty ball bearing
|Fixed displacement pump
|Total assembly failure rate
Ps = e(-500 x .000273376) = .872. This
value is probably pessimistic considering that much of the data is from
AVCO, however, it does provide a good starting point. Further, if
any single component is extremely critical, it is strongly recommended
that an in depth stress strength analysis is performed on that component.
Models of Mechanical Failure
Two of the more accurate models of mechanical failure are the maximum normal
stree theory and the distortion energy theory. [SADLON93]
These models rely on the design engineer being able to predict the level
of stress that will be applied to a component in order to estimate its
reliability. By using these models, first proposed at the turn of
the 20th century, an engineer can accurately estimate the reliability of
the component. Unfortunately, both of these methods requires a significant
amount of time, and if a generic data source can estimate that a component
will not fail within the mission time of the system with an acceptable
high probability and the component is not generally safety critical, then
this analysis can be avoided.
Maximum Normal Stress Theory
The maximum normal stress theory was first proposed by Rankine and is also
known as Rankine's theory of failure. [SALDON93]
The maximum normal stress theory is particularly well suited to materials
that are brittle as opposed to ductile materials. For example, materials
such as cast iron are well suited to this method while materials such as
urethane foam are not. Rankine's summary of this theory is that "failure
is predicted to occur in the multi-axial state of stress when the meximum
principal normal stress becomes equal to or exceeds the maximum normal
stress at the time of failure in a simple uniaxial stress test using a
specimen of the same material." This means that a sample of the same
material, cast iron for instance, is taken to a lab, and the iron is tested
for its maximum stress amounts. In order to do the testing, both
tensile, pulling, and compressive, squeezing, forces are applied to the
material in one axis only. When an engineer wants to determine wether
a part will fail or not, he can take multiaxial stress, create the normal
force vectors, and see if any of the normalized force vectors exceeds the
capability of the material. If a normalized stress does exceed the
capability of the material, the material is predicted to fail.
Distortion Energy Theory
Relationship to other topics
In order to understand mechanical reliability, it is necessary to understand
traditional reliability. The math behind calculating reliabilities
is the same. Furthermore, techniques, including parallel and serial
reliability are fully explained.
Traditional electronic reliability uses many of the same concepts of diversity
and redundancy in order to achieve reliability and fault tolerance.
There are many parallels between the two disciplines. There are however,
many differences. Mechanical parts wear due to corrosion, friction,
and other mechanical stresses. Electrical components do not wear
in the same manner, and so preparing for wear in the design phase is very
different. Electrical components experience drift, which the the
designer must make design accomodations. The effect of the different
types of wear are different approaches to maintenance. The mechanical
maintenance program may involve replacing parts that wear, while the electrical
approach may involve using feedback mechanisms and calibrating the electronics
to account of electrical wear, or drift.
Mechanical reliability is a very old craft practiced for more than a century.
The sciene behind mechanical reliability is constantly improving with better
lubricants and materials beign researched and created. The methods
to estimate and produce reliable mechanical components exist. Using
the existing tools and methods, extremely reliable mechanical systems can
be created. Just like in most forms of reliability, using diversity
and replication can create fault tolerant systems for safety critical systems.
Annotated Reference List
T. R. Moss and J. E. Strutt, "Data sources for reliability design analsys,"
in Proceedings of the Institution of Mechanical Engineers, vol.
27 Number E1, pp. 13-19, Institute of Mechanical Engineers, 1993.
A good list of data sources for reliability data of all types of mechanical components.
W. G. Ireson, C. F. Coombs, Jr., and R. Y. Moss, Handbook
of Reliability Engineering and Management. New York, New York: McGraw-Hill,
Good book on reliability engineering in general.
R. J. Sadlon, Mechanical Applications in Reliability
Engineering. Rome, New York: Reliability Analysis Center, 1993.
A good comprehensive introduction into mechanical reliability.
I. M. Hutchings, etc., New Directions in Tribology, Bury, Saint
Edmunds, UK, Institution of Mechanical Engineers, Mechanical Engineering
Publications, Limited, 1997.
Current research on tribology.
G. W. Stachowiak and A. W. Batchelor, Engineering Tribology. Amsterdam,
The Netherlands: Elsevier, 1993.
Good introductory book on tribology.
M. J. Neale, ed., Tribology Handbook. London: Butterworths, 1973.
Comprehensive, if dated, tribology reference.
J. A. Collins, Failure of Materials in Mechanical Design: Analysis,
Prediction, Prevention. New York, New York: John Wiley and Sons, Inc.,
second ed., 1993.
Textbook on why materials fail and the various types
of failure that materials can have.
Reliability Analysis Center http://rac.iitri.org/
Organization working for the US Air Force that publishes reliability data.
I need to complete the section on the distortion energy theory, I require
getting a more original source, instead of quoting quotes.
Go To Project